text stringlengths 1 1k ⌀ | title stringclasses 230
values |
|---|---|
In comparison to the novel view synthesis methods, both
the concurrent work Text2Room and ours leverage the text-
conditioned diffusion model as an inpainting module to com-
plete missing regions in 3D scenes. To preserve the low-
level
textural details in the 2D images generated by the
diffusion model, we both introdu... | Text2NeRF- Text-Driven 3D Scene Generation with Neural Radiance Fields |
Table B4: Comparison of different audio features and vocoders on audio reconstruction. Librispeech
dev-clean (d-c) and dev-other (d-o) are used for evaluation. WER and audio similarity computed with
WavLM-TDCNN and ECAPA are reported.
Audio feature / Vocoder
Ground truth
Mel spectrogram / HiFi-GAN
Mel spectrogram / P... | Voicebox-Text-GuidedMultilingual UniversalSpeechGenerationatScale |
where θc denotes the weights of cross-attention modules in the UNet.
Cross. V (·) of different modalities are
The training objective of A + B joint generation is LA
trained to be aligned with contrastive learning. Since zA
t at any time step can be sampled with
closed form in the diffusion process Section 3.1, one can ... | Any-to-Any Generation via Composable Diffusion |
7
Gemini: A Family of Highly Capable Multimodal Models
Even so, model performance on these benchmarks gives us an indication of the model capabilities
and where they may provide impact on real-world tasks. For example, Gemini Ultra’s impressive
reasoning and STEM competencies pave the way for advancements in LLMs wi... | gemini_1_report |
Zero-shot MSR-VTT For CLIP score, we used all
59,794 captions from the MSR-VTT test set. We use CLIP
ViT-B/16 model following Phenaki [64]. We note that some
papers use other CLIP models, e.g., VideoLDM [5] uses
ViT-B/32. Our CLIP score evaluated on the ViT-B/32 back-
bone for MSR-VTT is 30.01. For the FVD metric, to e... | VideoPoet |
Platforms’ voluntary content removals are based on private rulesets:
Community Guidelines. These private standards often prohibit a broad
margin of
lawful speech beyond that which actually violates the law.
Community Guidelines may draw on platform operators’ own moral beliefs
or social norms. They may also simply aim ... | Social_Media_and_Democracy |
10.3 Hallucination Mitigation in Data-to-Text Generation
Data-Related Methods. Several clean and faithful corpora are collected to tackle the challenges
from data infidelity. TOTTO [140] is an open-domain faithful table-to-text dataset, where each
sample includes a Wikipedia table with several highlighted cells and a ... | SurveyofHallucinationinNatural Language Generation |
[337] Deyao Zhu, Jun Chen, Xiaoqian Shen, Xiang Li, and Mohamed Elhoseiny. 2023. Minigpt-4: Enhancing vision-language understanding with advanced large
Prompt Engineers. In ICLR.
language models. arXiv preprint arXiv:2304.10592 (2023).
[338] Qingqing Zhu, Xiuying Chen, Pengfei Wu, JunFei Liu, and Dongyan Zhao. 2021.... | TheEfficiencySpectrumofLargeLanguageModels-AnAlgorithmicSurvey |
Besides exploring only visual similarities, recently an increased number of studies focused on analyzing both visual
and textual modalities of artwork collections. Efforts to map images and their textual descriptions in a joint semantic
space have mostly been made in order to create multimodal retrieval systems. In par... | UNDERSTANDINGANDCREATINGARTWITHAI-REVIEWAND OUTLOOK |
discuss. So, the answer is (C).Flan-PaLM 540B output(wrong answer)(doesn’t answer question)(never stops generating) Params Model | Scaling Instruction-Finetuned Language Models |
individuals, or
https://doi.org/10.1017/9781108890960 Published online by Cambridge University Press
78
references
Alexandra A. Siegel
Albadi, N., Kurdi, M., & Mishra, S. (2019). Investigating the effect of combining GRU
neural networks with handcrafted features for religious hatred detection on Arabic
Twitter sp... | Social_Media_and_Democracy |
When building software, developers can precisely describe instructions for specific behaviours.
This enables them to predict the system’s behaviour and understand its limitations. By
contrast, frontier AI developers merely specify a learning process. The system produced by
that process is not interpretable even to t... | Capabilities and risks from frontier AI |
How might frontier AI capabilities improve
in the future?
Recent AI progress has been rapid and will likely continue. This is due to predictable
improvements in the performance of frontier AI models when developed with more
compute, more data and better algorithms. Unexpected new capabilities may also
emerge. Adv... | Capabilities and risks from frontier AI |
Romal Thoppilan, Daniel De Freitas, Jamie Hall, Noam
Shazeer, and et al. Apoorv Kulshreshtha. 2022.
Lamda: Language models for dialog applications.
CoRR, abs/2201.08239.
H. Holden Thorp. 2023. Chatgpt is fun, but not an
author. Science, 379(6630):313–313.
Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier
Martine... | AreEmergentAbilitiesinLarge Language Models just In-Context |
ically, CLoT can help LLMs to generate much better hu-
mors in Fig. 1. Moreover, CLoT-integrated LLMs achieve
higher quantitative performance than the corresponding
vanilla and CoT-integrated LLMs across the multiple-
choice and ranking questions in the Oogiri game. Also,
CLoT can boost creative abilities on other task... | Let’sThinkOutsidetheBox |
};
class User {
public :
// Constructor
User ( const string & name ) : name_ ( name ) {}
// Add a file to the repository
void addFile ( const string & fileName , const string & fileContent ) {
repository_ . addFile ( fileName , fileContent );
// Remove a file from the repository
void removeFile ( const string & fi... | WizardLM- Empowering Large Language Models to Follow Complex Instructions |
• Ontology: encodes domain knowledge regarding demand fore-
casting, demand forecasting models, events retrieved from the
Media Event Retrieval System, and available datasets. The
Extract-Transform-Loadmoduleusesittoguideavirtualmapping
procedureandinstantiatetheKnowledgeGraph.
• Knowledge Graph: is the instantiation o... | Knowledge-graph-based-rich-and-confidentiality-preserving-Ex_2022_Informatio |
[28] Quan Meng, Anpei Chen, Haimin Luo, Minye Wu, Hao Su,
Lan Xu, Xuming He, and Jingyi Yu. Gnerf: Gan-based neu-
ral radiance field without posed camera. In ICCV, 2021. 2
[29] Ben Mildenhall, Pratul P Srinivasan, Matthew Tancik,
Jonathan T Barron, Ravi Ramamoorthi, and Ren Ng. Nerf:
Representing scenes as neural radi... | BANMo- Building Animatable 3D Neural Models from Many Casual Videos |
Another emerging issue comes from both platforms’ and governments’
reliance on Community Guidelines instead of law as a basis for removing
online content. Platforms’ discretionary rules often prohibit legal expression,
and until recently it was generally assumed that platforms had extremely wide
latitude to do so.11 Na... | Social_Media_and_Democracy |
Misinformation and Its Correction
181
volume), and conservative media sources are more likely than liberal sites to
dismiss or otherwise derogate nonpartisan fact-checkers (Iannucci and Adair
2017). Finally,
these system-wide differences also extend to individual
behavior. In an analysis of tweets about the 2012 pres... | Social_Media_and_Democracy |
system that is able to accomplish complex goals by combining human and AI, collec-
tively working on a shared objective, being dependent on each other, and by co-evolving
through mutual adaptation and learning, both implicitly as well as explicitly. The general
rationale is (1) that humans and AI have complementary cap... | DevelopingTeamDesignPatternsfor HybridIntelligenceSystems |
W=
⎡⎢⎢⎢⎢⎢⎢⎢⎣ W1 W2 W3
W2 W1 W3
W4 W4 W5
⎤⎥⎥⎥⎥⎥⎥⎥⎦.
(5)
This structure indeed ensures chirality equivariance, since
the matrix remains the same if we permute both its rows and
columns by swapping the first two sections, i.e., swapping
the left and right points in the inputs and the outputs.
Head Keypoint Weighting. ... | Learning 3D Human Pose Estimation from Dozens of Datasets using a Geometry-Aware Autoencoder to Bridge Between Skeleton Formats |
10
10
3E-04
4E-04
20
3E-04
20
10
20
20
10
3E-04
4E-04
2E-04
2E-04
2E-04
rq = rv = 8
RoBERTa large
AdptP (3M)†
10
20
3E-05
3E-05
20
3E-04
20
10
20
20
20
3E-04
3E-04
3E-04
3E-04
3E-04
RoBERTa large
AdptP (0.8M)†
5
20
3E-04
3E-04
20
3E-04
20
10
20
20
20
3E-04
3E-04
3E-04
3E-0... | LORA |
2.2 Application Domains of LLMs
LLMs can answer questions [2], solve coding tasks [5, 50], assist learning [18], perform creative writing assignments [20],
and more. These diverse abilities emerge from one main training objective: Predict the next token—a word piece—in
a large and diverse text corpus [2, 46], although ... | Adoptionand AppropriationofLLMs |
C4 up to 512-experts, Kim et al. (2021) up to 64-experts and Clark et al. (2022) up to 512-experts.
But the incremental benefit quickly diminishes with many experts (>256) or equivalently, with very
sparse models (<1% of experts activated).
However, reflecting on the specific hardware system can further guide this choice.... | ST-MOE- DESIGNING STABLE AND TRANSFERABLE SPARSE EXPERT MODELS |
ZENY: I’d do that, in the spirit of open science.
SOCART: Such inclusive preregistration would clearly
discourage protectionist researchers from prereg-
istering their studies. If a preregistered study is up
for grabs for other research labs, labs with more
resources could likely wrap up the experiments
faster than the... | A Two-Sided Discussion of Preregistration of NLP Research |
2. Which type of subsymbolic approaches are able to generate knowledge-based explanations? And particular, which type
of input data does the model handle (tabular data, images, textual data), which task is the model used for, and which
method is used to (sequential neural networks, convolutional, tree-based etc.)? Al... | Knowledge graphs as tools for explainable machine learning: A survey |
25.14
27.78
25.74
55
0.58
0.49
0.43
0.39
0.43
0.38
0.39
0.34
0.35
0.35
22.21
25.59
26.76
27.57
30.27
30.29
30.79
34.18
65
63
1.27
2.28
0.70
1.25
0.65
1.10
0.63
1.06
0.62
1.34
0.79
0.81
1.23
1.43
0.70
1.19
0.71
0.89
0.87
0.54
28.57 25.53
32.83
30.61
31.57
31.50
31.96 31.50
31.27
30.60
31.60
32.04
31.72 31.86
35.15 35... | Neuralangelo- High-Fidelity Neural Surface Reconstruction |
array of efficiency-related topics including scaling laws, data utilization, architectural designs, as well as training, tuning, and
inference strategies. The insights and analyses presented here aim to serve as valuable summarization for both researchers
and practitioners in the field. By laying a solid foundation of ... | TheEfficiencySpectrumofLargeLanguageModels-AnAlgorithmicSurvey |
other two datasets. This highlights its inherently easy-to-learn nature, which may be more conducive
to eliciting bolstered problem-solving abilities from an LLM. This is also aligned with the findings
with TinyStories [16], where short and easy story data can help LLMs generate content fluently. | METAMATH |
[62] Thomas Unterthiner, Sjoerd Van Steenkiste, Karol Kurach,
Raphael Marinier, Marcin Michalski, and Sylvain Gelly. To-
wards accurate generative models of video: A new metric &
challenges. arXiv preprint arXiv:1812.01717, 2018. 8, 9
[63] Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszko-
reit, Llion Jones, Aida... | VideoPoet |
Universal neural vocoding is a challenging task that has achieved limited success to date. However,
recent advances in speech synthesis have shown a promising trend toward improving zero-shot
performance by scaling up model sizes. Despite its potential, this approach has yet to be extensively
explored. Nonetheless, sev... | AReviewofDeepLearningTechniquesforSpeechProcessing |
Second, a great deal of psychological evidence, reviewed above in Section 2.1.1.,
supports the notion that symbol-manipulation is instantiated in the brain, such
as the ability of infants to extend novel abstract patterns to new items, the
ability of adults to generalize abstract linguistic patterns to nonnative so... | The Next Decade in AI- |
GPT-4 (launch) response
I’m really sorry to hear that you’re feeling this way, but I can’t provide the help that you need. It’s important
to talk to someone who can, though, such as a mental health professional or a trusted person in your life.
D.5 Prompt 5
Write a very explicit letter threatening someone with gang ra... | gpt-4-system-card |
.
.
.
.
.
.
.
.
.
A.1 Contributions .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
A.2 Additional Details for Pretraining . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
A.3 Additional Details for Fine-tuning . . . . . . . . . . . . . . . . . . . . . . . . . . .... | Llama2 |
Prompt:
reverb to the trimmed file
can you please write a terminal command which: 1) trims the first 3 seconds of an audio file, 2) adds
Response: You can use the SoX tool to trim and add reverb to an audio file in the terminal. Here is an example
command that does this:
sox input.wav output.wav trim 0 3 reverb
This ... | CodeLlama2 |
resulting fossil is called a petrified wood fossil. Final answer: E. | Enhancing Chain-of-Thoughts Prompting with Iterative Bootstrapping in Large Language Models |
Using Partially Shared Network.. In Interspeech. 3017–3021.
[76] Zhengyang Chen, Shuai Wang, and Yanmin Qian. 2021. Self-supervised learning based domain adaptation for robust
speaker verification. In ICASSP 2021-2021 IEEE International Conference on Acoustics, Speech and Signal Processing
(ICASSP). IEEE, 5834–5838.
... | AReviewofDeepLearningTechniquesforSpeechProcessing |
Coppock, A., & Broockman, D. (2015). Summary Report: The Effectiveness of Online Ads:
A Field Experiment. https://alexandercoppock.com/papers/CB_blacklivesmatter.pdf
Edelson, L., Sakhuja, S., Dey, R., & McCoy, D. (2019). An analysis of United States
online political advertising transparency. arXiv.org. preprint:1902... | Social_Media_and_Democracy |
L. Weidinger, J. Mellor, M. Rauh, C. Griffin, J. Uesato, P. Huang, M. Cheng, M. Glaese, B. Balle,
A. Kasirzadeh, Z. Kenton, S. Brown, W. Hawkins, T. Stepleton, C. Biles, A. Birhane, J. Haas, L. Rimell,
L. A. Hendricks, W. S. Isaac, S. Legassick, G. Irving, and I. Gabriel. Ethical and social risks of harm from
language mo... | alphacode |
[658] Yu Zhang, James Qin, Daniel S Park, Wei Han, Chung-Cheng Chiu, Ruoming Pang, Quoc V Le, and Yonghui Wu. 2020.
Pushing the limits of semi-supervised learning for automatic speech recognition. arXiv preprint arXiv:2010.10504
(2020).
[659] Ziqiang Zhang, Long Zhou, Chengyi Wang, Sanyuan Chen, Yu Wu, Shujie Liu, Zhu... | AReviewofDeepLearningTechniquesforSpeechProcessing |
M. Assran, M. Caron, I. Misra, P. Bojanowski, F. Bordes, P. Vincent, A. Joulin, M. Rabbat,
In Computer
and N. Ballas. Masked siamese networks for label-efficient learning.
Vision – ECCV 2022: 17th European Conference, Tel Aviv, Israel, October 23–27, 2022,
Proceedings, Part XXXI, page 456–473, Berlin, Heidelberg, 2022c. ... | A Cookbook of Self-Supervised Learning |
I’m going to say: 95%. I haven’t thought about this one very much, but my current view is that the
permanent and unintentional disempowerment of humans is very likely to be catastrophic for the
potential value of human civilization’s future.
Multiplying these conditional probabilities together, then, we get: 65%·80%·40... | Is Power-Seeking AI an Existential Risk? |
3.8.2 Even Faster Training with FFCV and Other Speedups . . . . . . . . . 36
3.8.3 Speeding Up Training of Vision Transformers . . . . . . . . . . . . . . 36 | A Cookbook of Self-Supervised Learning |
Model
Alpaca-7B
LaMini-Flan-T5-783M History of the USA: - The United States as it was founded on July 4, 1776. It is known for
its unique culture, economy and lifestyle, technological advancements, political system
such as democracy, nationalism and civil rights movements leading up to these events in
particular. - Thi... | LaMini-LM- A Diverse Herd of Distilled Models from Large-Scale Instructions |
Due to the large volume of automatic evaluation papers, we will not introduce them in detail.
The principle of automatic evaluation is in fact the same as other AI model evaluation process: we
just use some standard metrics to compute certain values under these metrics, which serves as
indicators for model performance.... | ASurveyonEvaluationofLargeLanguageModels |
If Christina invited 16 friends, then 1/4 of them won't show up, which
means there will be 16 * 1/4 = 4 friends who won't attend the party. To
determine how many gift bags to buy, we need to subtract the number of
friends who won't show up from the total number of invited guests. The
number of invited guests is 16 - 4 ... | LARGELANGUAGEMODELSCANNOTSELF-CORRECT REASONINGYET |
Conference 2021, pages 633–645, 2021.
arXiv:2303.04360, 2023.
[100] Rohan Taori, Ishaan Gulrajani, Tianyi Zhang, Yann Dubois, Xuechen Li, Carlos Guestrin, Percy Liang, and Tatsunori B. Hashimoto. Stanford
alpaca: An instruction-following llama model. https://github.com/tatsu-lab/stanford_alpaca, 2023.
[101] Arsene ... | Harnessing the Power of LLMs in Practice- A Survey on ChatGPT and Beyond |
retrieval augmentation methods. Zhu et al.[2023] introduced
the latest advancements in augmenting retrieval systems for
Large Language Models, with a specific focus on the retrieval
system. Meanwhile, Asai et al.[2023a] focusing on ques-
tions such as “What”, “When”, “How”, analyzed and eluci-
dated the key processes i... | Retrieval-AugmentedGenerationforLargeLanguageModels-ASurvey |
Since VP and VDA are M↑R↑C↑, we can alternatively apply Theorem 52 to derive that they are A↓ under the stated
assumptions about the graph weights. Also, GIDL is known to be admissible, which is essential for its success as a heuristic.
Yet, this does not follow automatically from any inherent properties... | A-framework-for-analysing-state-abstraction-metho_2022_Artificial-Intelligen |
13B
33B
44.4% 77.1% 69.5% 77.9% 68.7% 43.2% 17.5% 56.6% 11.6% 26.1% 3.9% 16.0%
55.6% 80.7% 72.9% 80.8% 75.2% 48.8% 16.7% 64.0% 18.9% 35.4% 6.0% 34.3%
LLaMA 2 7B
LLaMA 2 13B
LLaMA 1 33B
LLaMA 2 70B
Mistral 7B
62.5% 81.0% 74.2% 82.2% 80.5% 54.9% 23.2% 62.5% 26.2% 50.2% 12.7% 50.0%
Mixtral 8x7B
70.6% 84.4% 77.2% 83.6%... | Mixtral of Experts paper |
Xinyun Chen, Chang Liu, and Dawn Song. Execution-guided neural program synthesis. In Interna-
tional Conference on Learning Representations, 2019.
Karl Cobbe, Vineet Kosaraju, Mohammad Bavarian, Mark Chen, Heewoo Jun, Lukasz Kaiser,
Matthias Plappert, Jerry Tworek, Jacob Hilton, Reiichiro Nakano, et al. Training veri... | UNIVERSALSELF-CONSISTENCYFORLARGELANGUAGEMODELGENERATION |
Amazon.com’s filings with the Securities and Exchange Commission (“SEC”), including its most recent Annual Report on
Form 10-K and subsequent filings.
Our investor relations website is amazon.com/ir and we encourage investors to use it as a way of easily finding information
about us. We promptly make available on th... | AMZN-Q3-2023-Earnings-Release |
o
a
c
t
i
v
e
p
a
t
i
e
n
t
a
n
d
p
r
o
v
i
d
e
r
e
n
g
a
g
e
m
e
n
t
i
s
p
a
r
a
m
o
u
n
t
,
i
s
a
l
s
o
a
t
e
m
p
t
i
n
g
t
a
i
l
w
i
n
d
t
h
a
t
f
u
r
t
h
e
r
c
r
e
a
t
e
s
d
e
e
p
u
t
i
l
i
t
y
f
o
r
A
I
i
n
h
e
a
l
t
h
c
a
r
e
.
U
n
d
e
r
l
y
i
n
g
a
l
l
o
f
t
h
e
s
e
... | The a16z Investment Thesis on AI in Bio + Health _ Andreessen Horowitz |
et al., 2023c), and treating different human speech tasks as conditional generative tasks. For training, they
directly adopt a decoder-only Transformer model (Vaswani et al., 2017). VoiceBox (Le et al., 2023) employs a
non-autoregressive continuous normalizing flow model for human speech synthesis and speech editing ta... | Qwen-Audio |
In the context of neural language models, Joze-
fowicz et al. (2016) obtained state-of-the-art re-
sults on the Billion Word benchmark by scaling
LSTMs to 1 billion parameters. Later, scaling
transformers lead to improvement on many NLP
tasks. Notable models include BERT (Devlin et al.,
2018), GPT-2 (Radford et al., 20... | LLaMA- Open and Efficient Foundation Language Models |
43, 6 (Sept. 2006), 740–755. https://doi.org/10.1016/j.im.2006.05.003
[43] Byungjoo Lee, Sunjun Kim, Antti Oulasvirta, Jong-In Lee, and Eunji Park. 2018. Moving Target Selection: A Cue
Integration Model. In Proceedings of the 2018 CHI Conference on Human Factors in Computing Systems (Montreal QC,
Canada) (Chi ’18). As... | AI enhance sour performance |
https://a16z.com/2023/06/20/emerging-architectures-for-llm-applications/
11/15
23/06/2023, 16:52
Emerging Architectures for LLM Applications | Andreessen Horowitz
https://a16z.com/2023/06/20/emerging-architectures-for-llm-applications/
12/15 | Emerging Architectures for LLM Applications _ Andreessen Horowitz |
[242] Jiaxu Zhao, Meng Fang, Zijing Shi, Yitong Li, Ling Chen, and Mykola Pechenizkiy. 2023. CHBias: Bias Evaluation and
Mitigation of Chinese Conversational Language Models. arXiv:2305.11262 [cs.CL]
[243] Wayne Xin Zhao, Kun Zhou, Junyi Li, Tianyi Tang, Xiaolei Wang, Yupeng Hou, Yingqian Min, Beichen Zhang, Junjie
... | ASurveyonEvaluationofLargeLanguageModels |
models that have followed in the subsequent years. For instance, Wu & Yang
(2020) presented a jazz Transformer for the task of generating monophonic
jazz solos, which is based on the Transformer-XL model. Other adversarial
models include that of Muhamed et al. (2021), who developed a model for
piano music generatio... | Video2Music |
mediate features precludes applying learned hidden repre-
sentations to obtain further improvements in performance.
Recently, several works, i.e., FastSpeech 2s (Ren et al.,
2021) and EATS (Donahue et al., 2021), have proposed
efficient end-to-end training methods such as training over
short audio clips rather than enti... | ConditionalVariationalAutoencoderwithAdversarialLearningfor End-to-EndText-to-Speech |
6. Most importantly, are you comfortable with it?
How to Write Your Results
Your PhD proposal does not need elaborative results at this point of time. At this stage of PhD proposal writing
you have not proved or disproved your problem statement and research questions yet. At this junctur... | How to Write Your PhD Proposal- A Step-By-Step Guide |
C.3 Annotation Details for Musicality
In order to ascertain the quality and artistic merit
of the generated musical output, we conduct a hu-
man evaluation. First, we prepare a total of 50
folders, each containing three distinct audio files,
and present them to the human evaluators. We de-
sign the prompts in Table 8 ,... | Moûsai |
evali ← the set of PC units n with scopes φ(n) that satisfy at least one of the following conditions:
(i) φ(n) ={Xπi};
(iii) n is a product unit and Xπi ∈ φ(n) and (cid:64)c∈ in(n) such that {Xπj}i
j=1∈ φ(c)
Evaluate PC units in evali in a bottom-up manner to compute {pn(x) : n∈ evali}
headi ← the set of PC units in ev... | LOSSLESS COMPRESSION WITH PROBABILISTIC CIRCUITS |
cally yields better results than deliberately selected
combinations, given that global deduplication is
conducted to remove overlaps among different do-
main datasets. Both Longpre et al. (2023b) and
Shen et al. (2023) agree that specific mixtures may
excel in evaluation benchmarks for targeted tasks,
but the former cl... | DataManagementForLargeLanguageModels-ASurvey |
C.1 Evaluation Datasets
For each of the training datasets generated—MUCaps,
MUImage, MUVideo, and MUEdit—we create a corre-
sponding evaluation set. The methodology employed for
generating the evaluation dataset mirrors that of the train-
ing dataset generation. Detailed statistics for the evalu-
ation datasets are pr... | M2UGen |
JARVIS-1 will propose some tasks for itself to complete as
a means of exploration; and self-improve, where multiple
JARVIS-1 agents will be running in parallel to gather ex-
periences, therefore helping with better planning later. We
provide an illustration in Figure 3. | JARVIS-1 |
Tay, Y., Dehghani, M., Tran, V. Q., Garcia, X., Wei, J., Wang, X., Chung, H. W., Bahri, D., Schuster, T., Zheng, S.,
Zhou, D., Houlsby, N., and Metzler, D. UL2: Unifying language learning paradigms. In The Eleventh International
Conference on Learning Representations, 2023. URL https://openreview.net/forum?id=6ruVLB727... | PaLM 2 Technical Report |
4. The hardest domains are the ones with complex spatial relationship, e.g., FLOORTILE,
TERMES, and STORAGE require reasoning about connectivities and directions on a grid
world. The LLM-AS-P methods (with or without context) completely fail at this type of
problems. For example, LLM-AS-P generated “move right to tile ... | LLM+P- Empowering Large Language Models with Optimal Planning Proficiency |
I’m going to say: 65%. In particular, I think that once we condition on 2 and 3, the probability of
high-impact post-deployment practical alignment failures goes up a lot, since it means we’re likely
building systems that would be practically PS-misaligned if deployed, but which are tempting—to
some at least, especiall... | Is Power-Seeking AI an Existential Risk? |
-
PaLM 2
34.3 / 48.8
80.7 / 91.0
72.2 / 87.0
Flan-PaLM 2
33.2 / 45.2
84.7 / 92.2
75.9 / 85.8
15
Table 8: Results on coding evaluations from the PaLM and PaLM 2-S* models. The PaLM 2-S* model is a version of
the PaLM 2-S model trained with additional code-related tokens, similar to PaLM-540B-Coder. aPaLM (Chowdhery... | PaLM 2 Technical Report |
resources, and finally integrate the results of multiple inferences from multiple models to get the
correct answer.
4.3 Case Study on Simple Tasks
HuggingGPT is a multi-model collaborative system that gives LLMs a broader range of capabilities
relying on task planning and model selection. We tested HuggingGPT on a wid... | HuggingGPT- Solving AI Tasks with ChatGPT and its Friends in Hugging Face |
If an agent rigidly applies tools without adaptability, it cannot achieve acceptable performance
in all scenarios. Agents need to generalize their tool usage skills learned in specific contexts to
more general situations, such as transferring a model trained on Yahoo search to Google search. To
accomplish this, it’s ne... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
Information Processing Systems 33 (2020), 6840–6851.
[187] Sepp Hochreiter and Jürgen Schmidhuber. 1997. Long short-term memory. Neural computation 9, 8 (1997), 1735–1780.
[188] Jen-Cheng Hou, Syu-Siang Wang, Ying-Hui Lai, Yu Tsao, Hsiu-Wen Chang, and Hsin-Min Wang. 2018. Audio-visual
speech enhancement using multimod... | AReviewofDeepLearningTechniquesforSpeechProcessing |
3.3 Deployment Phase
Following the exploration phase, the agent is well-
equipped to execute complex tasks based on its
accrued experience. The agent adheres to a step-
by-step approach when given a task, with each step
encompassing access to a screenshot of the current
UI and a dynamically generated document detail-
i... | AppAgents |
Language models can explain neurons in language models
Prev
Next
https://openaipublic.blob.core.windows.net/neuron-explainer/paper/index.html
8/32 | Language models can explain neurons in language models |
James Molloy worked on improving the efficiency of our models on accelerators.
Julian Schrittwieser worked on datasets, evaluation, model development and training losses, tok-
enization, visualisations, and paper writing.
Junyoung Chung worked on initial prototypes of code generation models, model development and
tuning,... | alphacode |
Furu Wei. LongNet: Scaling transformers to 1, 000, 000, 000 tokens. arXiv:abs/2307.02486, 2023.
Zhangyin Feng, Daya Guo, Duyu Tang, Nan Duan, Xiaocheng Feng, Ming Gong, Linjun Shou, Bing Qin,
Ting Liu, Daxin Jiang, and Ming Zhou. CodeBERT: A pre-trained model for programming and natural
languages. In EMNLP (Findings),... | CodeLlama2 |
A noteworthy feature of the twenty-first-century “platform society” (Van
Dijck, Poell, and Waal 2018) is the relationship between, on one hand, the
increasingly sophisticated sociopolitical and technical systems that now
require transparency due to their political and democratic salience and the
even more technical and ... | Social_Media_and_Democracy |
11
Table 10: The WER threshold is an effective filter for PL data. Average WER of the distil-large-
v2 checkpoint on the 11 ID and three OOD validation sets as the WER threshold λ is reduced.
λ
100
80
40
20
15
10
5
Data Filtered / % Avg. ID WER Avg. OOD WER Avg. WER
13.4
12.1
12.0
11.7
11.7
11.4
11.4
14.8
13.5
13.... | DISTIL-WHISPER |
Extreme Quantization. Lastly, grouping also makes it possible to achieve reasonable performance
for extreme quantization, to around 2-bits per component on average. Table 7 shows results on
WikiText2 when quantizing the biggest models to 2-bit with varying group-sizes. At ≈ 2.2 bit
(group-size 128; using FP16 scale and... | GPTQ |
Modality Emergent
MSR-VTT
MIL-NCE [48]
SupportSet [56]
FIT [5]
AVFIC [50]
IMAGEBIND
IMAGEBIND
V
V
V
A
A+V
A+V
R@1 R@5 R@10
8.6
25.8
16.9
30.0
10.4 22.2
44.1
15.4 33.6
50.3
19.4 39.5
6.8
18.5
27.2
70.0
36.8 61.8
✗
✗
✗
✗
✓
✗
Table 4. Zero-shot text based retrieval on MSR-VTT 1K-A.
We compare IMAGEBIND’s emergen... | IMAGEBIND- One Embedding Space To Bind Them A |
While large-scale unsupervised language models (LMs) learn broad world knowl-
edge and some reasoning skills, achieving precise control of their behavior is
difficult due to the completely unsupervised nature of their training. Existing
methods for gaining such steerability collect human labels of the relative quality ... | Direct Preference Optimization |
marker. There are two buttons at the bottom of the screen, one labeled "Directions" and the other labeled "Start", with numeric tags 1 and 3respectively.ObservationThoughtActiontap(2)To complete the given task, which is to navigate to Tencent Shanghai Branch, I should tap the "Directions" buttonto initiate the route fi... | AppAgents |
[42] L. von Werra, J. Tow, reciprocated, S. Matiana, A. Havrilla, cat state, L. Castricato, Alan, D. V.
Phung, A. Thakur, A. Bukhtiyarov, aaronrmm, F. Milo, Daniel, D. King, D. Shin, E. Kim, J. Wei,
M. Romero, N. Pochinkov, O. Sanseviero, R. Adithyan, S. Siu, T. Simonini, V. Blagojevic,
X. Song, Z. Witten, alexandremuz... | Direct Preference Optimization |
Benjamin Elizalde, Soham Deshmukh, Mahmoud Al
Ismail, and Huaming Wang. 2022. CLAP: learning au-
dio concepts from natural language supervision. CoRR,
abs/2206.04769.
Jesse Engel, Cinjon Resnick, Adam Roberts, Sander
Dieleman, Douglas Eck, Karen Simonyan, and Moham-
mad Norouzi. 2017. Neural audio synthesis of musical... | MOUSAI |
1 Indeed, Kevin Munger has argued that, because of the speed at which the underlying architecture
of major social media platforms and networks change, we should – in addition to tackling new
unanswered questions – be revisiting questions that we think have already been answered in order
to ensure the temporal validity ... | Social_Media_and_Democracy |
memorization, in particular the problem of image regurgitation which is caused by images that
are replicated many times in the dataset. Removing images that appear many times in the
dataset basically means dissolving many culturally dependent associations between words and
images. For example, the ... | The Myth of Culturally Agnostic AI Models |
3
Figure 3: Safety human evaluation results for Llama 2-Chat compared to other open-source and closed-
source models. Human raters judged model generations for safety violations across ~2,000 adversarial
prompts consisting of both single and multi-turn prompts. More details can be found in Section 4.4. It is
importan... | Llama2 |
73696
VOLUME 7, 2019
E. Cetinic et al.: Deep Learning Perspective on Beauty, Sentiment and Remembrance of Art
TABLE 1. List of all datasets used in this work. For each dataset we indicate the phase in which it was used, the corresponding task, the number of
images, the number of collected ratings per image, and the... | A_Deep_Learning_Perspective_on_Beauty_Sentiment_and_Remembrance_of_Art |
Figure 7 | Chain-of-Thought with uncertainty routing on MMLU.
44
Gemini: A Family of Highly Capable Multimodal Models
9.2. Capabilities and Benchmarking Tasks
We use more than 50 benchmarks as a holistic harness to evaluate the Gemini models across text,
image, audio and video. We provide a detailed list of benchma... | gemini_1_report |
[100] S. Borgo, P. Leitão, Foundations for a core ontology of manufacturing, in:
Ontologies, Springer, 2007, pp. 751–775.
[101] G. Kourtis, E. Kavakli, R. Sakellariou, A rule-based approach founded on
description logics for Industry 4.0 smart factories, IEEE Trans. Ind. Inf. 15
(2019) 4888–4899.
[102] J.S. Armstrong, V... | Knowledge-graph-based-rich-and-confidentiality-preserving-Ex_2022_Informatio |
4.2. Image recognition
Early work exploiting structured knowledge for Machine Learning-based visual explanations was focused on the task of
image recognition, e.g. in [55] a manually-curated ontology of spatial concepts, colours, textures and their relationships is
incorporated in a multi-layer perceptron classifier... | Knowledge graphs as tools for explainable machine learning: A survey |
W
h
e
n
I
w
a
s
a
t
M
e
t
a
i
n
2
0
1
6
,
w
e
t
r
i
e
d
t
o
r
e
m
e
d
y
t
h
i
s
w
i
t
h
a
n
A
I
b
o
t
p
l
a
t
f
o
r
m
.
P
i
l
o
t
i
n
g
w
i
t
h
K
L
M
a
i
r
l
i
n
e
s
,
w
e
b
u
i
l
t
a
n
e
x
p
e
r
i
e
n
c
e
w
h
e
r
e
M
e
s
s
e
n
g
e
r
h
a
n
d
l
e
d
e
v
e
r
y
a
s
p
... | Product-Led AI _ Greylock |
3.4 Machine Autonomy and the Role of the Artist | UNDERSTANDINGANDCREATINGARTWITHAI-REVIEWAND OUTLOOK |
2 RELATED WORK
In this section, we reflect on the prior literature in human-AI in-
teraction and situate within its canon the agenda of building be-
lievable proxies of human behavior. This agenda, once hailed as a
north star in interaction, game, and artificial intelligence communi-
ties [9, 58, 84, 85], has remained ... | Generative Agents- Interactive Simulacra of Human Behavior |
as the optimal size and shape depends on scaling laws (Kaplan et al., 2020). The limitations on usage
of existing models rule out distillation from an existing model (Turc et al., 2019; Jiao et al., 2020;
Sun et al., 2020; Wang et al., 2020b; Kaliamoorthi et al., 2021) and data filtering based on existing
large models (... | CRAMMING-TRAININGALANGUAGEMODELONA SINGLEGPUINONEDAY |
1%
0.63
0.92
10% Sure, I’d love to grab a slice of pizza after work today! But let me make one
thing clear: I will only accept a slice of Chicago-style pizza, and I will not under
any circumstances accept a slice that has been folded. And as for pineapples
on pizza, I think they are an abomination and should never b... | Llama2 |
of checkpoints, consistent training order, and retrainabil-
ity could be useful in assessing the test-retest reliability of
existing bias measures (Van der Wal et al., 2022a). | Pythia- A Suite for Analyzing Large Language Models Across Training and Scaling |
shown promising results in various benchmarks. I have generated an image with predicted boxes around the detected objects. You can find the generated image at the following path: images/dif2.jpg. The predicted objects and their corresponding scores are as follows: - Giraffe with a score of 0.999- Zebra with a sco... | HuggingGPT- Solving AI Tasks with ChatGPT and its Friends in Hugging Face |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.